AM5 and Hybrid mode.

[AsRock]

Well i thought i give this a try as the last time i tried it was back when Lucid? did it around intel Ivy lake and later, back then i got all kinds of issue's like the ones you get with laptops and such were the GPU would not kick in on demand. all so i believe they tried charging people too for newer drivers and basically gave up bothering.

So i thought i give it a shot again as you can do this with the new AM5 platform even though there did not seem at 1st glance a reason too, how ever i like to see some people with AM5 and the newer RX79XX series and using multi monitor to try this see if you can get around the higher power usages.

I did not expect it to work as well as it does tbh and thought it be worth a little time investigating as it seems like no ones really bothering. As i am using a 6950XT power usages are reported in adrenalin is 5w which is about 2w difference (5w).

How ever checking my UPS while the system is idle the system idles around 90w which is about 30-37w less (normally about 117-127).

I have tried a bunch of games and had the odd quirk but starting to seem very plausible.

Power usage of GPU seems to be the GPU usage, as if it was not my UPS would pick up on what looks like extra usage.

 

[Launcestonian]

Necro regenerate this thread... lol..
This concept here is intriguing to say the least, I notice this option in the bios of my Asus B650m TUF Gaming board with the 8600G. I was considering an experiment with an 8700G & 7900 XTX together working in tandem with their graphics power (iGPU + dGPU) combined as one force on my X670E board. Knowing full well that the 8700G is cache crippled compared to 7000 & 9000 series but the FCLK on 8000 series OCs significantly more than what any 7000 or 9000 series can possible do due to its monolithic design, so in theory the half reduced cache could be less of a bottle neck than expected - 2500 FCLK is not out of the question here & fully stable in Aida 64 stability test + MTP testing as well.
If gaming at 1440p is the standard, CPU IPC is a thing but obviously not as much as at 1080p, however from my experience with OC on the 8600G it is possible to get another 200 - 250MHz static OC out of this series as even with PBO enhanced with full options for it, I only ever see a maximum of 150MHz on my 8600G & that is with thermals well in check & significantly below 95C at all times. Sometimes I think the PBO algorithm applied to 8000 series is not applied in the best way compared to what I have seen on my 7600X & 9700X - but then again, it could be the boards their on that are playing a part here as well but I digress for the moment.
I'm not sure if anyone here at TPU has had experience with Hybrid graphics on AM5 apart from the OP, there is this acknowledgment from AMD about it working some time ago already but with RDNA 2 platform. Microsoft also talk about it here, however it is with laptop systems only they mention, but the architecture of 8000 series is fundamentally for laptops anyway so there's that.
Just putting this out there to get some feedback... thanks!

 

[tabascosauz]

Necro regenerate this thread... lol..
This concept here is intriguing to say the least, I notice this option in the bios of my Asus B650m TUF Gaming board with the 8600G. I was considering an experiment with an 8700G & 7900 XTX together working in tandem with their graphics power (iGPU + dGPU) combined as one force on my X670E board. Knowing full well that the 8700G is cache crippled compared to 7000 & 9000 series but the FCLK on 8000 series OCs significantly more than what any 7000 or 9000 series can possible do due to its monolithic design, so in theory the half reduced cache could be less of a bottle neck than expected - 2500 FCLK is not out of the question here & fully stable in Aida 64 stability test + MTP testing as well.
If gaming at 1440p is the standard, CPU IPC is a thing but obviously not as much as at 1080p, however from my experience with OC on the 8600G it is possible to get another 200 - 250MHz static OC out of this series as even with PBO enhanced with full options for it, I only ever see a maximum of 150MHz on my 8600G & that is with thermals well in check & significantly below 95C at all times. Sometimes I think the PBO algorithm applied to 8000 series is not applied in the best way compared to what I have seen on my 7600X & 9700X - but then again, it could be the boards their on that are playing a part here as well but I digress for the moment.
I'm not sure if anyone here at TPU has had experience with Hybrid graphics on AM5 apart from the OP, there is this acknowledgment from AMD about it working some time ago already but with RDNA 2 platform. Microsoft also talk about it here, however it is with laptop systems only they mention, but the architecture of 8000 series is fundamentally for laptops anyway so there's that.
Just putting this out there to get some feedback... thanks!

I don't think the hybrid GPU concept is for adding performance like you are suggesting (like how Bulldozer APU iGPUs could Hybrid Crossfire with dGPU). iirc Hybrid Crossfire also had strict GPU choice limitations. Mainly just 2 things:

• If running a monitor off the iGPU can get you to an important breakpoint, ie. your dGPU can step down to a lower VRAM Pstate and save idle power
• Regardless of which screen is connected to what, you can Win+Shift+arrowkeys the game window around onto any screen (even one connected to iGPU) and enjoy dGPU performance there.

On performance it starts to fall apart a bit. It's either high FCLK or UCLK that the iGPU doesn't really seem to like, and can start occasionally/rampantly flickering depending on the screen/resolution/Hz/connection. Might be a bit better on yours, monolithic APU.

And using the iGPU does, as expected, negatively affect VSOC requirements and max UCLK (maybe FCLK??) that can be attained stable. Which is not a surprise as iGPUs have always been a drag on uncore in Ryzen CPUs and GPUs.

The Microsoft article on their tech is just referring to their own (native) alternative to AMD and Nvidia's GPU switching tech.

 

[AusWolf]

Necro regenerate this thread... lol..
This concept here is intriguing to say the least, I notice this option in the bios of my Asus B650m TUF Gaming board with the 8600G. I was considering an experiment with an 8700G & 7900 XTX together working in tandem with their graphics power (iGPU + dGPU) combined as one force on my X670E board. Knowing full well that the 8700G is cache crippled compared to 7000 & 9000 series but the FCLK on 8000 series OCs significantly more than what any 7000 or 9000 series can possible do due to its monolithic design, so in theory the half reduced cache could be less of a bottle neck than expected - 2500 FCLK is not out of the question here & fully stable in Aida 64 stability test + MTP testing as well.
If gaming at 1440p is the standard, CPU IPC is a thing but obviously not as much as at 1080p, however from my experience with OC on the 8600G it is possible to get another 200 - 250MHz static OC out of this series as even with PBO enhanced with full options for it, I only ever see a maximum of 150MHz on my 8600G & that is with thermals well in check & significantly below 95C at all times. Sometimes I think the PBO algorithm applied to 8000 series is not applied in the best way compared to what I have seen on my 7600X & 9700X - but then again, it could be the boards their on that are playing a part here as well but I digress for the moment.
I'm not sure if anyone here at TPU has had experience with Hybrid graphics on AM5 apart from the OP, there is this acknowledgment from AMD about it working some time ago already but with RDNA 2 platform. Microsoft also talk about it here, however it is with laptop systems only they mention, but the architecture of 8000 series is fundamentally for laptops anyway so there's that.
Just putting this out there to get some feedback... thanks!

It sounds all fancy, but does a higher FCLK actually contribute to your real-world experience?

My two cents on hybrid graphics:

• As @tabascosauz said, it doesn't increase performance. You can choose which GPU to run X application on, and that's it. It's not meant as a replacement for Crossfire (which is totally 100% dead).
• You can connect your display to the iGPU, or the dGPU, and choose to run X app on one or the other.
  • However, this does work flawlessly with Nvidia, but it doesn't with AMD. I have a HTPC with an i7-11700 and a 1050 Ti in it, but I've got the Intel Xe iGPU connected to the TV because it gives a better image. Games and 3D apps in general run on the 1050 Ti which transmits the signal to the TV through the iGPU without a problem. So the iGPU serves as an extra display output.
  • On AMD, if you connect the monitor to the iGPU and run your games on the dGPU, you'll experience weird hitches and random drops in performance that can't be explained.
    • What I find the best solution is connecting your main screen to the dGPU, and any secondaries to the iGPU. That keeps your dGPU idle usage down while also allowing you to connect multiple displays. You can still run any app on the iGPU if you choose to, while keeping the dGPU free for gaming on your main display.

 

[Launcestonian]

I don't think the hybrid GPU concept is for adding performance like you are suggesting (like how Bulldozer APU iGPUs could Hybrid Crossfire with dGPU). iirc Hybrid Crossfire also had strict GPU choice limitations. Mainly just 2 things:

• If running a monitor off the iGPU can get you to an important breakpoint, ie. your dGPU can step down to a lower VRAM Pstate and save idle power
• Regardless of which screen is connected to what, you can Win+Shift+arrowkeys the game window around onto any screen (even one connected to iGPU) and enjoy dGPU performance there.

On performance it starts to fall apart a bit. It's either high FCLK or UCLK that the iGPU doesn't really seem to like, and can start occasionally/rampantly flickering depending on the screen/resolution/Hz/connection. Might be a bit better on yours, monolithic APU.

And using the iGPU does, as expected, negatively affect VSOC requirements and max UCLK (maybe FCLK??) that can be attained stable. Which is not a surprise as iGPUs have always been a drag on uncore in Ryzen CPUs and GPUs.

The Microsoft article on their tech is just referring to their own (native) alternative to AMD and Nvidia's GPU switching tech.

If you do a net search, you'll find numerous comments & discussions in enthusiasts hangouts like reddit for example, where they did not see performance gains unless the DP cable is plugged into the DP output on the back of the motherboard, so the whole graphics subsystem is routed out through that part apparently is what I saw. But I suppose it depends in the greater scheme of things what search engine you use to find this information... just like practically every other topic out there in the world today.
Have a lapped 7600X here running 1:1 & can do 6400CL32 no problems with iGPU as only gpu in whole system, I have benched it ( & even stress tested it in 3DMarks game tests - fully stable & passes just fine) & achieved legendary status in some 3D mark runs already, this on B650 board too & 1440p monitor. So the maturing of AGESA has helped a lot here imo. VSOC is only very slightly higher than VSOC on my 9700X for a given FLCK frequency value. The Microsoft article is relevant because you can't have OS support for hybrid graphics unless the OS is co-operating with that tech in the first place & even then only on latter versions of win 11 ( which will be a pickle for those who hate win 11)

It sounds all fancy, but does a higher FCLK actually contribute to your real-world experience?

My two cents on hybrid graphics:

• As @tabascosauz said, it doesn't increase performance. You can choose which GPU to run X application on, and that's it. It's not meant as a replacement for Crossfire (which is totally 100% dead).
• You can connect your display to the iGPU, or the dGPU, and choose to run X app on one or the other.
  • However, this does work flawlessly with Nvidia, but it doesn't with AMD. I have a HTPC with an i7-11700 and a 1050 Ti in it, but I've got the Intel Xe iGPU connected to the TV because it gives a better image. Games and 3D apps in general run on the 1050 Ti which transmits the signal to the TV through the iGPU without a problem. So the iGPU serves as an extra display output.
  • On AMD, if you connect the monitor to the iGPU and run your games on the dGPU, you'll experience weird hitches and random drops in performance that can't be explained.
    • What I find the best solution is connecting your main screen to the dGPU, and any secondaries to the iGPU. That keeps your dGPU idle usage down while also allowing you to connect multiple displays. You can still run any app on the iGPU if you choose to, while keeping the dGPU free for gaming on your main display.

Higher & stable FCLK aids in overcoming the Zen 4 & 5 platforms biggest hinderance to system RAM performance. 2500FCLK on my 8600G substantially beats my 9700X with 2200FCLK in Aida64 benching & this with 12 thread to 16 thread as well! No one realises how the FCLK is the key to better system memory performance until they've tinkered with an 8000 series APU on a decent board with decent RAM & then compared to the 7000 & 9000 series.
I have no interest in running multiple displays as trendy as that is today. My concern is solely with gaming performance on one monitor running 1440p x 3440p & high refresh rates (165Hz) only. Fiddling around with TVs bores me.

 

[tabascosauz]

If you do a net search, you'll find numerous comments & discussions in enthusiasts hangouts like reddit for example, where they did not see performance gains unless the DP cable is plugged into the DP output on the back of the motherboard, so the whole graphics subsystem is routed out through that part apparently is what I saw. But I suppose it depends in the greater scheme of things what search engine you use to find this information... just like practically every other topic out there in the world today.
Have a lapped 7600X here running 1:1 & can do 6400CL32 no problems with iGPU as only gpu in whole system, I have benched it ( & even stress tested it in 3DMarks game tests - fully stable & passes just fine) & achieved legendary status in some 3D mark runs already, this on B650 board too & 1440p monitor. So the maturing of AGESA has helped a lot here imo. VSOC is only very slightly higher than VSOC on my 9700X for a given FLCK frequency value. The Microsoft article is relevant because you can't have OS support for hybrid graphics unless the OS is co-operating with that tech in the first place & even then only on latter versions of win 11 ( which will be a pickle for those who hate win 11)


Higher & stable FCLK aids in overcoming the Zen 4 & 5 platforms biggest hinderance to system RAM performance. 2500FCLK on my 8600G substantially beats my 9700X with 2200FCLK in Aida64 benching & this with 12 thread to 16 thread as well! No one realises how the FCLK is the key to better system memory performance until they've tinkered with an 8000 series APU on a decent board with decent RAM & then compared to the 7000 & 9000 series.
I have no interest in running multiple displays as trendy as that is today. My concern is solely with gaming performance on one monitor running 1440p x 3440p & high refresh rates (165Hz) only. Fiddling around with TVs bores me.

Sure, except my point stands. Hybrid graphics on desktop isn't Hybrid Crossfire. I can't say I've heard anyone claim believably to the contrary (not that Reddit is usually some kind of authoritative source) that the RDNA2 iGPU functions as any sort of Crossfire-esque performance adder.

Hybrid graphics works fine with Nvidia on mobo HDMI-out, so clearly DP isn't required, although some people did say that flickering is minimized on a DP connection. Unless you again are referring to the idea of iGPU adding performance.

CASO is nice and all and I wouldn't be surprised if it has continued to exist in the driver, but I think we all know by now that features don't exactly translate seamlessly like that between laptops and desktops. AMD has never advertised anything similar on the deskop, and there isn't as much of a need for it. In any case, AMD hasn't really made much of a dent in the mobile dGPU market after that generation of hardware in which CASO was announced (6000 Rembrandt + RX6000 RDNA2), so it seems logical that it hasn't had much traction since. I can't say I've ever noticed 680M enhancing the RX 6700S' performance in my G14 on any driver version. And that's supposed to be CASO's intended use case.

As for 16MB vs 32MB, Phoenix APUs are nice but AMD knows what they are doing by halving L3. The current state of APUs vs. chiplets is not really changed from any of the past Ryzen generations. You can (like in Renoir vs. Matisse, like in Cezanne vs. Vermeer) clock the APUs to the moon and they will spank the other CPUs in synthetic tests, especially AIDA. That still never indicates anything about gaming performance, no reasonable amount of mem/FCLK/CPU OC manages to overcome the cache deficit of not having 32MB L3.

Strix APUs may be one to watch out for when they come to desktop due to the increase in L3, but seeing as last I checked the L3 seems to be reverting to Zen 2 design (no longer a unified block in 1 CCX, 16+8MB instead of 24MB), so it may be more of the same in the end.

 

[DirtyDingusMcgee]

I had a phenom II rig i ran hybrid crossfire on. 890gx (so radeon 4290) paired with an xfx 5570. it provided a decent boost on many titles (i was playing a lot of DDO and Quake wars at the time, and it helped them both), but some things ran much worse.....and emulators didn't like it at all. A neat idea. I do think the modern implementation of having an apu and a gpu, and the ability to switch between them on command, is a more elegant solution.

 

[kapone32]

The only thing with the IGPU on 7000 CPUs is that it is weak. It does not even support 60hz. I do see the IGPU used when I am playing some Games though.

 

[Launcestonian]

The only thing with the IGPU on 7000 CPUs is that it is weak. It does not even support 60hz. I do see the IGPU used when I am playing some Games though.

Its intended for corporate use, where 2D is all that's needed. I use it on one of my desktops all the time, no problems ever with web browsing, image editing, etc..

 

[AsRock]

The only thing with the IGPU on 7000 CPUs is that it is weak. It does not even support 60hz. I do see the IGPU used when I am playing some Games though.

You sure it don't support 60Hz as it seems like some thing i would of noticed.

 

[Taraquin]

Partially relevant: With my 4070 and 7800x3d I get 10W lower idle consumption if I enable igpu even if I dont use it at all. Using monitorsoftware I get idle voltage on cores of down to 0.47v and consumption if package of 11-15W vs 0.69v and 15-22W idle voltage/consumption if I disable igpu in bios. From wall wattage w/o monitor is 47-49W igpu enabled vs 57-60W igpu disabled.

 

[Launcestonian]

Sure, except my point stands. Hybrid graphics on desktop isn't Hybrid Crossfire. I can't say I've heard anyone claim believably to the contrary (not that Reddit is usually some kind of authoritative source) that the RDNA2 iGPU functions as any sort of Crossfire-esque performance adder.

Hybrid graphics works fine with Nvidia on mobo HDMI-out, so clearly DP isn't required, although some people did say that flickering is minimized on a DP connection. Unless you again are referring to the idea of iGPU adding performance.

CASO is nice and all and I wouldn't be surprised if it has continued to exist in the driver, but I think we all know by now that features don't exactly translate seamlessly like that between laptops and desktops. AMD has never advertised anything similar on the deskop, and there isn't as much of a need for it. In any case, AMD hasn't really made much of a dent in the mobile dGPU market after that generation of hardware in which CASO was announced (6000 Rembrandt + RX6000 RDNA2), so it seems logical that it hasn't had much traction since. I can't say I've ever noticed 680M enhancing the RX 6700S' performance in my G14 on any driver version. And that's supposed to be CASO's intended use case.

As for 16MB vs 32MB, Phoenix APUs are nice but AMD knows what they are doing by halving L3. The current state of APUs vs. chiplets is not really changed from any of the past Ryzen generations. You can (like in Renoir vs. Matisse, like in Cezanne vs. Vermeer) clock the APUs to the moon and they will spank the other CPUs in synthetic tests, especially AIDA. That still never indicates anything about gaming performance, no reasonable amount of mem/FCLK/CPU OC manages to overcome the cache deficit of not having 32MB L3.

Strix APUs may be one to watch out for when they come to desktop due to the increase in L3, but seeing as last I checked the L3 seems to be reverting to Zen 2 design (no longer a unified block in 1 CCX, 16+8MB instead of 24MB), so it may be more of the same in the end.

I've never mentioned Hybrid crossfire & that's not the title of the original thread. Hybrid graphics is not 2 dGPU working togeather. It's the iGPU in an APU working with 1 dGPU, that's all it is. Yes I am referring to the iGPU adding performance, that's the whole point of hybrid graphics.

 

[kapone32]

You sure it don't support 60Hz as it seems like some thing i would of noticed.

Maybe but when I plugged it into my 4K monitor that was what the settings were and there was no option to change.

 

[tabascosauz]

I've never mentioned Hybrid crossfire & that's not the title of the original thread. Hybrid graphics is not 2 dGPU working togeather. It's the iGPU in an APU working with 1 dGPU, that's all it is. Yes I am referring to the iGPU adding performance, that's the whole point of hybrid graphics.

Getting extra perf from iGPU to benefit the dGPU is literally just what CF used to do. Which is also what you are referring to.

Like I said before, AMD never advertised or intended its newer hybrid graphics conception to benefit performance like you are describing. It would be nice if you could link to those other users' testing showing that it can, otherwise, that's not what hybrid graphics is.

Perhaps 8700G has some obscure abilities or features that carry over from it being mobile silicon, like it's Cezanne predecessors, but as far as hybrid graphics is advertised and debuted with AM5 chiplet CPUs specifically, performance benefits have not been part of the picture.

Partially relevant: With my 4070 and 7800x3d I get 10W lower idle consumption if I enable igpu even if I dont use it at all. Using monitorsoftware I get idle voltage on cores of down to 0.47v and consumption if package of 11-15W vs 0.69v and 15-22W idle voltage/consumption if I disable igpu in bios. From wall wattage w/o monitor is 47-49W igpu enabled vs 57-60W igpu disabled.

Sounds like an interesting board or AGESA bug? I've had iGPU disabled at one point but it made no difference so it went right back.

 

[Taraquin]

Getting extra perf from iGPU to benefit the dGPU is literally just what CF used to do. Which is also what you are referring to.

Like I said before, AMD never advertised or intended its newer hybrid graphics conception to benefit performance like you are describing. It would be nice if you could link to those other users' testing showing that it can, otherwise, that's not what hybrid graphics is.

Perhaps 8700G has some obscure abilities or features that carry over from it being mobile silicon, like it's Cezanne predecessors, but as far as hybrid graphics is advertised and debuted with AM5 chiplet CPUs specifically, performance benefits have not been part of the picture.



Sounds like an interesting board or AGESA bug? I've had iGPU disabled at one point but it made no difference so it went right back.

I dunno why, but if 'works' for unknown reasons